code/interface/annotators_v5.py

import gradio as gr
import json
import os
from datetime import datetime

# --- PATH CONFIGURATION ---
# DATA_PATH = "/home/mshahidul/readctrl/data/synthetic_dataset_diff_labels/syn_data_with_gs_summary_en_0_20.json"
DATA_PATH = "/home/mshahidul/readctrl/data/synthetic_dataset_diff_labels/syn_data_diff_labels_bn_0_80.json"
SAVE_ROOT = "/home/mshahidul/readctrl/data/annotators_validate_data_Bangla_(0_80)"
os.makedirs(SAVE_ROOT, exist_ok=True)

# --- UI HTML COMPONENTS (Kept same as original) ---
GUIDE_HTML = """
<div style="background-color: #f9f9f9; padding: 15px; border-left: 6px solid #4CAF50; border-radius: 4px; margin-bottom: 20px;">
    <h3>Rating Guide: Medical Text Difficulty</h3>
    <table style="width:100%; border-collapse: collapse; text-align: left;">
        <tr style="background-color: #e8f5e9;">
            <th style="padding: 8px; border: 1px solid #ddd;">Score</th>
            <th style="padding: 8px; border: 1px solid #ddd;">Description</th>
        </tr>
        <tr><td><b>1</b></td><td><b>Very Easy:</b> Simple words, no medical jargon.</td></tr>
        <tr><td><b>2</b></td><td><b>Easy:</b> Conversational medical terms.</td></tr>
        <tr><td><b>3</b></td><td><b>Moderate:</b> Standard patient education material.</td></tr>
        <tr><td><b>4</b></td><td><b>Hard:</b> Significant technical jargon.</td></tr>
        <tr><td><b>5</b></td><td><b>Very Hard:</b> Specialist-level / Academic.</td></tr>
    </table>
</div>
"""

EXAMPLES_HTML = """
<div style="background-color: #ffffff; padding: 15px; border: 1px solid #ddd; border-radius: 4px;">
    <h3 style="color: #2e7d32;">Reference Examples</h3>
    <div style="display: flex; gap: 15px;">
        <div style="flex: 1; background-color: #f1f8e9; padding: 10px; border-radius: 4px;">
            <h4>Level 1-2</h4>
            <p>"She had a kidney problem... a big blood clot blocked veins in her brain."</p>
        </div>
        <div style="flex: 1; background-color: #ffebee; padding: 10px; border-radius: 4px;">
            <h4>Level 4-5</h4>
            <p>"Idiopathic NS inaugurated by cerebral venous thrombosis extended to the right jugular vein."</p>
        </div>
    </div>
</div>
"""
def parse_diff_label_texts(raw_value):
    """
    Parse diff_label_texts that may be:
    - dict (already parsed)
    - JSON string
    - Python-dict-like string (single quotes)
    """
    if isinstance(raw_value, dict):
        return raw_value

    if not isinstance(raw_value, str):
        return {}

    text = raw_value.strip()
    if not text:
        return {}

    # Prefer strict JSON first; fall back to Python literal parsing.
    try:
        parsed = json.loads(text)
        return parsed if isinstance(parsed, dict) else {}
    except json.JSONDecodeError:
        pass

    try:
        parsed = ast.literal_eval(text)
        return parsed if isinstance(parsed, dict) else {}
    except (ValueError, SyntaxError):
        return {}
import ast
# --- DATA LOADING ---
def normalize_dataset(raw_dataset):
    """
    Normalize different dataset layouts into a flat queue where each item has:
    index, id, label, generated_summary.
    """
    normalized = []

    for item in raw_dataset:
 

        # New layout: {"diff_label_texts": {label: text, ...}}
        diff_label_texts = item.get("diff_label_texts")
        if isinstance(diff_label_texts, dict):
            for label, text in diff_label_texts.items():
                normalized.append({
                    "index": item.get("index"),
                    "id": item.get("id"),
                    "label": label,
                    "generated_summary": text
                })
            
        else:
            diff_label_texts = parse_diff_label_texts(item.get("diff_label_texts"))
            for label, text in diff_label_texts.items():
                normalized.append({
                    "index": item.get("index"),
                    "id": item.get("id"),
                    "label": label,
                    "generated_summary": text
                })
            


    return normalized


if os.path.exists(DATA_PATH):
    with open(DATA_PATH, "r", encoding="utf-8") as f:
        RAW_DATASET = json.load(f)
        FULL_DATASET = normalize_dataset(RAW_DATASET)
        print(len(FULL_DATASET))
    assert FULL_DATASET, f"No valid items found in dataset: {DATA_PATH}"
else:
    assert False, f"Data file not found at {DATA_PATH}"

# --- PERSISTENCE HELPERS ---
def get_user_dir(username):
    clean_username = "".join([c for c in username if c.isalnum() or c in (' ', '_', '-')]).strip() or "anonymous"
    return os.path.join(SAVE_ROOT, clean_username)

def save_state(user_dir, state_dict):
    with open(os.path.join(user_dir, "state.json"), "w") as f:
        json.dump(state_dict, f, indent=4)

def load_state(user_dir):
    state_path = os.path.join(user_dir, "state.json")
    if os.path.exists(state_path):
        with open(state_path, "r") as f:
            return json.load(f)
    return None

# --- LOGIC FUNCTIONS ---
def get_current_ui_values(state):
    """Helper to get UI values for the current index, including previous ratings if they exist."""
    idx = state['current_index']
    current_item = state['queue'][idx]
    
    # Check if we already have a rating for this specific index
    existing_rating = 3  # Default
    for res in state['results']:
        if res['queue_position'] == idx:
            existing_rating = res['rating']
            break
            
    progress = f"Item {idx + 1} of {len(state['queue'])}"
    return current_item['generated_summary'], progress, existing_rating

def start_session(username):
    if not username:
        gr.Warning("Please enter a username!")
        return [gr.update()] * 5

    user_dir = get_user_dir(username)
    os.makedirs(user_dir, exist_ok=True)
    existing_state = load_state(user_dir)
    
    if existing_state:
        gr.Info(f"Welcome back! Resuming from item {existing_state['current_index'] + 1}.")
        state = existing_state
    else:
        state = {
            "username": username,
            "current_index": 0,
            "queue": list(FULL_DATASET),
            "results": [],
            "completed": False
        }
        save_state(user_dir, state)

    text, progress, rating = get_current_ui_values(state)
    return (gr.update(visible=False), gr.update(visible=True), text, progress, rating, state)

def submit_rating(doc_slider, state):
    if state is None: return "", "Error", 3, 3, None

    user_dir = get_user_dir(state['username'])
    idx = state['current_index']
    current_item = state['queue'][idx]
    
    # Update existing rating if editing, otherwise append
    new_result = {
        "queue_position": idx,
        "index": current_item.get('index', idx),
        "doc_id": current_item.get('id', current_item.get('index', 'no_id')),
        "label": current_item.get('label', 'no_label'),
        "rating": doc_slider,
        "timestamp": datetime.now().strftime("%Y-%m-%d %H:%M:%S")
    }
    
    # Logic to overwrite existing rating for this index
    state['results'] = [r for r in state['results'] if r['queue_position'] != idx]
    state['results'].append(new_result)
    state['results'].sort(key=lambda x: x['queue_position']) # Keep sorted

    if idx + 1 < len(state['queue']):
        state['current_index'] += 1
        save_state(user_dir, state)
        # Save results file
        with open(os.path.join(user_dir, "annotation_results.json"), "w") as f:
            json.dump(state['results'], f, indent=4)
            
        text, progress, rating = get_current_ui_values(state)
        return text, progress, rating, state
    else:
        state['completed'] = True
        save_state(user_dir, state)
        return "✅ ALL TASKS COMPLETED", "Status: Finished", 1, state

def go_back(state):
    if state is None or state['current_index'] <= 0:
        gr.Warning("Already at the first item.")
        return [gr.update()] * 3 + [state]

    state['current_index'] -= 1
    text, progress, rating = get_current_ui_values(state)
    return text, progress, rating, state

# --- UI INTERFACE ---
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    session_state = gr.State()

    gr.Markdown("# Medical Text Readability Annotation")
    
    with gr.Accordion("Instructions & Calibration", open=False):
        gr.HTML(GUIDE_HTML)
        gr.HTML(EXAMPLES_HTML)

    with gr.Column(visible=True) as intro_box:
        username_input = gr.Textbox(label="Enter Your Name/ID", placeholder="e.g., user_101")
        btn_start = gr.Button("Start / Resume Annotation", variant="primary")

    with gr.Column(visible=False) as task_box:
        progress_label = gr.Label(label="Overall Progress")
        doc_display = gr.Textbox(interactive=False, lines=12, label="Medical Text")
        doc_slider = gr.Slider(1, 5, step=1, label="Difficulty (1=Easy, 5=Hard)", value=3)
        
        with gr.Row():
            btn_prev = gr.Button("⬅️ Previous", variant="secondary")
            btn_submit = gr.Button("Submit & Next ➡️", variant="primary")

    # --- EVENT HANDLERS ---
    btn_start.click(
        fn=start_session, 
        inputs=[username_input],
        outputs=[intro_box, task_box, doc_display, progress_label, doc_slider, session_state]
    )
    
    btn_submit.click(
        fn=submit_rating,
        inputs=[doc_slider, session_state],
        outputs=[doc_display, progress_label, doc_slider, session_state]
    )

    btn_prev.click(
        fn=go_back,
        inputs=[session_state],
        outputs=[doc_display, progress_label, doc_slider, session_state]
    )

if __name__ == "__main__":
    demo.launch(share=True)